McNemar’s test has been carried out in this study to measure the significance between the classification of SVM and CART, SVM and RF and CART and RF. The 2 x 2 contingency table as tabulated in Table 6 was used to calculate the p-values.
1247
[Table 6 near here]
The null hypothesis of this test states that the probability of Test 1 being correctly classified is equal to the probability of Test 2 being correctly classified. Also, the probability of Test 1 being incorrectly classified is equal to the probability of Test 2 being incorrectly classified. In other words, Pa + Pb = Pa + Pc or Pb + Pd = Pc + Pd, which leads to Pb = Pc.
Pa = Probability of Test 1 being positive and Test 2 being positive Pb = Probability of Test 1 being positive and Test 2 being negative Pc = Probability of Test 1 being negative and Test 2 being positive Pd = Probability of Test 1 being negative and Test 2 being negative
The p-value will be calculated and the value of p < 0.05 is considered as a significant result, thus rejecting the null hypothesis. In this study, calculations of the p-value using the formula demonstrated by Foody (2004) were conducted for all the algorithms and the results are tabulated in Table 8.
[Table 7 near here]
The p-value obtained when comparing between SVM and RF is 0.28 (p > 0.05), while the other two comparisons obtained values of p < 0.05. Due to the robustness and powerful machine learning algorithms, SVM and RF algorithms can classify the pixels well. Hence, the comparison between SVM and RF gave a non-significant p-value > 0.05 and thus, accepted the null hypothesis.
5. Conclusion
In this study, we utilised 30 m Landsat data in the GEE platform to produce oil palm land cover maps over Peninsular Malaysia. The GEE platform is controllable and it provides
1305
options especially in selecting the processing methods, algorithms and data input.
Furthermore, it allows users to design the workflow based on their needs. In this study, three machine learning algorithms were used and the hyperparameters were tuned. Accuracy assessments for the classified maps were conducted using high-resolution Google Earth images and the map provided by the DOA. The comparison of the classified oil palm areas with the inventory provided by MPOB has shown that there is a large uncertainty of oil palm land cover in Perlis, Kedah and Selangor. Overall, CART, SVM and RF were able to classify the land cover maps and produced acceptable results by producing an overall accuracy of 80.08%, 93.16% and 86.50% respectively. Then, McNemar’s test was conducted and it showed that significant p-values were obtained when comparing CART to both SVM and RF.
However, the test showed a non-significant value when comparing between RF and SVM.
This shows that both methods can reliably be used to produce high accuracy maps in GEE and later be used to classify other crops. Moving on, such timely and high accuracy estimates of oil palm areas could be embedded with other ancillary GIS data for a variety of monitoring and decision-making applications, including yield prediction, supply-chain logistics,
commodity markets, bioenergy estimation and more.
GEE provides various geospatial data including Sentinel 2, Sentinel 1 and MODIS.
The utilisation of higher spatial resolution data such as Sentinel 2 with 20 m to 10 m of pixel size can be tested to improve the classification. Moreover, Sentinel 1 works with active sensors, and it is suitable to be used on tropical regions. The integration of Sentinel 1 data in the GEE platform can reduce the time needed to process huge amounts of radar data. On top of that, there are many more methods available in GEE to pre-process remote sensing data, in which some methods might produce good results and able to improve the accuracy of the
1365
data. In addition, the programmable platform produces the possibilities for the cloud computing GEE to be integrated with the powerful deep learning methods.
Acknowledgements
We would like to thank Universiti Putra Malaysia for their facilities and support for this research.
Sponsorship from the Engineering and Physical Sciences Research Council UK (EPSRC/RCUK) (Grant Number: EP/P018165/1- Newton Fund) is gratefully acknowledged. The comments from the anonymous reviewers in improving this article are highly appreciated.
References
Afonja T. 2017. Kernel functions. [accessed 2019 March 12].
https://towardsdatascience.com/kernel-function-6f1d2be6091.
Belgiu M, Drăguţ L. 2016. Random forest in remote sensing: A review of applications and future directions. ISPRS J Photogramm Remote Sens. 114:24-31.
doi:10.1016/j.isprsjprs.2016.01.011.
Bittencourt H, Clarke RT. 2003. Use of classification and regression trees (CART) to classify remotely-sensed digital images. In: IGARSS 2003. Proceedings of the IEEE Trans Geosci Remote Sens Symposium 2003. Toulouse, France, July 21–25: IEEE IGARSS.
p. 3751–3753.
Brid RS. 2018. Decision trees. Medium; [accessed 2019 January 20].
https://medium.com/greyatom/decision-trees-a-simple-way-to-visualize-a-decision-dc506a403aeb.
Bro-Jørgensen J, Brown ME, Pettorelli N. 2008. Using the satellite-derived normalized difference vegetation index (NDVI) to explain ranging patterns in a lek-breeding
1423
antelope: the importance of scale. Oecolgia. 158(1):177-182. doi:10.1007/s00442-008-1121-z
Calbury. 2016. Object-based Classification: Classification and Regression Tree (CART).
[Accessed 2019 March 3]. wiki.landscapetoolbox.org/doku.php/remote_sensing_
methods:classification_and_regression_tree_cart.
Camacho A, Correa CV, Arguello H. 2019. An analysis of spectral variability in hyperspectral imagery: a case study of stressed oil palm detection in Colombia. Int J Remote Sens.
40(19):7603-7623. doi:10.1080/01431161.2019.1595210.
Chander G, Markham BL, Helder DL. 2009. Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sens Environ. 113(5):893-903. doi:10.1016/j.rse.2009.01.007.
Charters LJ, Aplin P, Marston CG, Padfield R, Rengasamy N, Bin Dahalan MP, Evers S. 2019.
Peat swamp forest conservation withstands pervasive land conversion to oil palm plantation in North Selangor, Malaysia. Int J Remote Sens.40(19):1-30.
doi:10.1080/01431161.2019.1574996.
Cheng Y, Yu L, Cracknell AP, Gong P. 2016. Oil palm mapping using Landsat and PALSAR:
A case study in Malaysia. Int J Remote Sens. 37(22):5431-5442.
doi:10.1080/01431161.2016.1241448.
Cheng Y, Yu L, Xu Y, Lu H, Cracknell AP, Kanniah K, Gong P. 2018. Mapping oil palm extent in Malaysia using ALOS-2 PALSAR-2 data. Int J Remote Sens. 39(2):432-452.
doi:10.1080/01431161.2017.1387309.
Chong KL, Kanniah KD, Pohl C, Tan KP. 2017. A review of remote sensing applications for oil palm studies. Geo-spatial Information Science. 20(2):184-200.
doi:10.1080/10095020.2017.1337317.
De Alban, J, Connette G, Oswald P, Webb E. (2018). Combined Landsat and L-band SAR data improves land cover classification and change detection in dynamic tropical landscapes. Remote Sensing. 10(2):306. doi:10.3390/rs10020306.
Dong J, Xiao X, Menarguez MA, Zhang G, Qin Y, Thau D, Biradar C, Moore B. 2016.
Mapping paddy rice planting area in northeastern Asia with Landsat 8 images, phenology-based algorithm and Google Earth Engine. Remote Sens Environ. 185:142-154. doi:10.1016/j.rse.2016.02.016.
Duro DC, Franklin SE, Dubé, MG. 2012. A comparison of pixel-based and object-based image analysis with selected machine learning algorithms for the classification of agricultural landscapes using SPOT-5 HRG imagery. Remote Sens Environ. 118:259-272.
doi:10.1016/j.rse.2011.11.020.
Fahmi Z, Samah BA, Abdullah H. 2013. Paddy industry and paddy farmers well-being: a success recipe for agriculture industry in Malaysia. Asian Soc Sci. 9(3):177-181.
doi:10.5539/ass.v9n3p177.
Fawcett D, Azlan B, Hill TC, Kho LK, Bennie J, Anderson K. 2019. Unmanned aerial vehicle (UAV) derived structure-from-motion photogrammetry point clouds for oil palm (Elaeis guineensis) canopy segmentation and height estimation. Int J Remote Sens. 40(19):1-23.
doi:10.1080/01431161.2019.1591651.
Fitzherbert EB, Struebig MJ, Morel A, Danielsen F, Brühl CA, Donald PF, Phalan B. 2008.
How will oil palm expansion affect biodiversity?. Trends Ecol Evol. 23(10):538-545.
doi:10.1016/j.tree.2008.06.012.
Gambo J, Shafri HZM, Shaharum NSN, Abidin FAZ, Rahman MTA. 2018. Monitoring and Predicting Land Use-Land Cover (LULC) Changes Within and Around Krau Wildlife Reserve (KWR) Protected Area in Malaysia Using Multi-Temporal Landsat Data.
Geoplanning: journal of geomatics and planning. 5(1):23-44.
doi:10.14710/geoplanning.5.1.17-34.
Gislason PO, Benediktsson JA, Sveinsson JR. 2006. Random forests for land cover classification. Pattern Recognit Lett. 27(4):294-300. doi:10.1016/j.patrec.2005.08.011.
Glinskis EA, Gutiérrez-Vélez VH. 2019. Quantifying and understanding land cover changes by large and small oil palm expansion regimes in the Peruvian Amazon. Land Use Policy. 80:95-106. doi:10.1016/j.landusepol.2018.09.032.
Goldblatt R, You W, Hanson G, Khandelwal A. 2016. Detecting the boundaries of urban areas in india: A dataset for pixel-based image classification in google earth engine. Remote sensing. 8(8):634. doi:10.3390/rs8080634.
Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. 2017. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens Environ. 202:18-27. doi:/10.1016/j.rse.2017.06.031.
Gutiérrez-Vélez VH, DeFries R. 2013. Annual multi-resolution detection of land cover conversion to oil palm in the Peruvian Amazon. Remote Sens Environ. 129:154-167.
doi:10.1016/j.rse.2012.10.033.
Gupta O, Das AJ, Hellerstein J, Raskar R. 2018. Machine learning approaches for large scale classification of produce. Sci Rep. 8(1):5226-5233.
1601
Ivanovic B. 2016. Cross-validation and decision trees. [accessed 2019 January 15].
https://www.cs.utoronto.ca/~fidler/teaching/2015/slides/CSC411/tutorial3_CrossVal-DTs.pdf.
Jiang L, Wang W, Yang X, Xie N, Cheng Y. 2010. Classification methods of remote sensing image based on decision tree technologies. Paper presented at CCTA 2010. International Conference on Computer and Computing Technologies in Agriculture; Oct 20–25;
Nanchang, China.
Joshi N, Baumann M, Ehammer A, Fensholt R, Grogan K, Hostert P, Jepsen MR, Kuemmerie T, Meyfroidt P, Mitchard ETA, Reiche J, Ryan CM, Waske B. (2016). A review of the application of optical and radar remote sensing data fusion to land use mapping and monitoring. Remote Sensing. 8(1):70. doi:10.3390/rs8010070.
Jusoff K, Pathan M. 2009. Mapping of individual oil palm trees using airborne hyperspectral sensing: an overview. Applied physics research. 1(1):15-30.
Lee JSH, Wich S, Widayati A, Koh LP. 2016. Detecting industrial oil palm plantations on Landsat images with Google Earth Engine. Remote sensing applications: society and environment. 4:219-224. doi:10.1016/j.rsase.2016.11.003.
Li L, Dong J, Tengku SN, Xiao X. 2015. Mapping oil palm plantations in Cameroon using PALSAR 50-m orthorectified mosaic images. Remote sensing. 7(2):1206-1224.
doi:10.3390/rs70201206.
Li W, Fu H, Yu L, Cracknell A. (2016). Deep learning based oil palm tree detection and counting for high-resolution remote sensing images. Remote Sensing. 9(1):22.
doi:https://doi.org/10.3390/rs9010022.
Mahidin MU. 2018. Selected Agricultural Indicators Malaysia. Malaysia: Department of Statistics Malaysia; [accessed 2019 March 15].
https://www.dosm.gov.my/v1/index.php?r=column/cthemeByCat&cat=72&bul_id=Uj YxeDNkZ0xOUjhFeHpna20wUUJOUT09&menu_id=Z0VTZGU1UHBUT1VJMFlpa XRRR0xpdz09.
Maxwell AE, Warner TA, Fang F. 2018. Implementation of machine-learning classification in remote sensing: An applied review. Int J Remote Sens. 39(9):2784-2817.
doi:10.1080/01431161.2018.1433343.
Miettinen J, Shi C, Tan WJ, Liew SC. 2012. 2010 land cover map of insular Southeast Asia in 250-m spatial resolution. Pattern Recognit Lett. 3(1):11-20.
doi:10.1080/01431161.2010.526971.
Molnar C. 2016. Interpretable Machine Learning: A Guide for Making Black Box Models Explainable. 1st ed. Christoph Molnar. [accessed 2019 January 20].
https://christophm.github.io/interpretable-ml-book/tree.html.
Morel AC, Fisher JB, Malhi Y. 2012. Evaluating the potential to monitor aboveground biomass in forest and oil palm in Sabah, Malaysia, for 2000–2008 with Landsat ETM+ and ALOS-PALSAR. Int J Remote Sens. 33(11):3614-3639. doi:10.1080/01431161.2011.631949.
Nambiappan B, Ismail A, Hashim N, Ismail N, Shahari DN, Idris NAN, Omar N, Salleh KM, Hassan NAM, Din AK. 2018. Malaysia: 100 years of resilient palm oil economic performance. J. Oil Palm Res. 30(1):13-25. doi:10.21894/jopr.2018.0014.
Ng WPQ, Lam HL, Ng FY, Kamal M, Lim JHE. 2012. Waste-to-wealth: green potential from palm biomass in Malaysia. J Clean Prod. 34:57-65. doi:10.1016/j.jclepro.2012.04.004.
1719
Noi PT, Kappas M. 2018. Comparison of random forest, k-nearest neighbor, and support vector machine classifiers for land cover classification using Sentinel-2 imagery. Sensors.
18(1):1-20. doi:10.3390/s18010018.
Nooni IK, Duker AA, Van Duren I, Addae-Wireko L, Osei Jnr EM. 2014. Support vector machine to map oil palm in a heterogeneous environment. Int J Remote Sens.
35(13):4778-4794. doi:10.1080/01431161.2014.930201.
Oliphant AJ, Thenkabail PS, Teluguntla P, Xiong J, Gumma MK, Congalton RG, Yadav K.
2019. Mapping cropland extent of Southeast and Northeast Asia using multi-year time-series Landsat 30-m data using a random forest classifier on the Google Earth Engine Cloud. Int J Appl Earth Obs Geoinf. 81:110-124. doi:10.1016/j.jag.2018.11.014.
Pal M. 2005. Random forest classifier for remote sensing classification. Int J Remote Sens.
26(1):217-222. doi:10.1080/01431160412331269698.
Pal M, Maxwell AE, Warner TA. 2013. Kernel-based extreme learning machine for remote-sensing image classification. Pattern Recognit Lett. 4(9):853-862.
doi:10.1080/2150704X.2013.805279.
Patel NN, Angiuli E, Gamba P, Gaughan A, Lisini G, Stevens FR, Tatem AJ, Trianni G. 2015.
Multitemporal settlement and population mapping from Landsat using Google Earth Engine. Int J Appl Earth Obs Geoinf. 35:199-208. doi:10.1016/j.jag.2014.09.005.
Patel S. 2017. Chapter 2: SVM (Support Vector Machine) — Theory. [accessed 2019 March 10]. https://medium.com/machine-learning-101/chapter-2-svm-support-vector-machine-theory-f0812effc72.
Roy DP, Wulder MA, Loveland TR, Woodcock CE, Allen RG, Anderson MC, Helder D, Irons JR, Johnson DM, Kennedy R, et al. 2014. Landsat-8: Science and product vision for
terrestrial global change research. Remote Sens Environ. 145:154-172.
doi:10.1016/j.rse.2014.02.001.
Shafri HZM, Hamdan N. 2009. Hyperspectral imagery for mapping disease infection in oil palm plantationusing vegetation indices and red edge techniques. Am J Appl Sci.
6(6):1031-1035.
Shafri HZM, Anuar MI, Seman IA, Noor NM. 2011. Spectral discrimination of healthy and Ganoderma-infected oil palms from hyperspectral data. Int J Remote Sens. 32(22):7111-7129. doi:10.1080/01431161.2010.519003.
Shahar FM. 2016 March 18. Heat wave emergency if temperature exceeds 40 degrees Celsius for more than 7 days. New Straits Times. [accessed 2019 January 20].
https://www.nst.com.my/news/2016/03/133534/heat-wave-emergency-if-temperature-exceeds-40-degrees-celcius-more-7-days.
Shaharum NSN, Shafri HZM, Gambo J, Abidin FAZ. 2018. Mapping of Krau Wildlife Reserve
(KWR) protected area using Landsat 8 and supervised classification algorithms. Remote Sensing Applications: Society and Environment. 10:24-35.
doi:10.1016/j.rsase.2018.01.002.
Shaharum NSN, Shafri HZM, Ghani WAWAK, Samsatli S, Prince HM, Yusuf B, Hamud AM.
2019. Mapping the spatial distribution and changes of oil palm land cover using an open source cloud-based mapping platform. Int J Remote Sens. 40(19):1-18.
doi:10.1080/01431161.2019.1597311.
Shelestov A, Lavreniuk M, Kussul N, Novikov A, Skakun S. 2017. Exploring Google earth engine platform for Big Data Processing: Classification of multi-temporal satellite imagery for crop mapping. Front Earth Sci. 5(17):1-10. doi:10.3389/feart.2017.00017.
1837
Shuit SH, Tan KT, Lee KT, Kamaruddin AH. 2009. Oil palm biomass as a sustainable energy source: A Malaysian case study. Energy. 34(9):1225-1235.
doi:10.1016/j.energy.2009.05.008.
Sidhu N, Pebesma E, Câmara G. 2018. Using Google Earth Engine to detect land cover change: Singapore as a use case. Eur J Remote Sens. 51(1):486-500.
doi:10.1080/22797254.2018.1451782.
Thenkabail PS, Stucky N, Griscom BW, Ashton MS, Diels J, Van Der Meer B, Enclona E.
2004. Biomass estimations and carbon stock calculations in the oil palm plantations of African derived savannas using IKONOS data. Int J Remote Sens. 25(23):5447-5472.
doi:10.1080/01431160412331291279.
Yu X, Hyyppä J, Litkey P, Kaartinen H, Vastaranta M, Holopainen M. 2017. Single-sensor solution to tree species classification using multispectral airborne laser scanning. Remote Sensing. 9(2):108-123. doi:10.3390/rs9020108.
1895
Figure 2. The Earth Engine Javascript API.
Figure 4. (a) High-resolution Google Earth image, (b) Landsat 8 image.
Figure 6. The division of the tree in CART.
Figure 8. Subsamples in cross validation.
Figure 9. Classified oil palm land cover maps of Peninsular Malaysia, (a) CART, (b) RF and (c) SVM.
(c)
Figure 10. (a) High-resolution Google Earth image, (b) CART, (c) RF and (d) SVM.
(a) (b)
(c) (d)
Legend
:
Figure 11. (a) High-resolution Google Earth image, (b) CART, (c) RF and (d) SVM.
(a) (b)
(d) (c)
Legend
:
Name Description Pixel size (m) Wavelength (μm)
Band 1 Coastal aerosol 30 0.435 - 0.451
Band 2 Blue 30 0.452 - 0.512
Band 3 Green 30 0.533 - 0.590
Band 4 Red 30 0.636 - 0.673
Band 5 Near Infrared 30 0.851 - 0.879
Band 6 Short-wave Infrared 1 30 1.566 - 1.651
Band 7 Short-wave Infrared 2 30 2.107 - 2.294
Table 2. Additional layer to be included for classification.
Name Formula Reference/Source
NDVI 𝑁𝐼𝑅 ‒ 𝑅𝑒𝑑
𝑁𝐼𝑅 + 𝑅𝑒𝑑
(Bannari et al., 1995;
Maselli, 2004)
NDWI 𝐺𝑟𝑒𝑒𝑛 ‒ 𝑁𝐼𝑅
𝐺𝑟𝑒𝑒𝑛 + 𝑁𝐼𝑅 (Xu et al., 2010)
Blue Red 𝐵𝑙𝑢𝑒 ‒ 𝑅𝑒𝑑
Blue Green 𝐵𝑙𝑢𝑒 ‒ 𝐺𝑟𝑒𝑒𝑛 (Murray et al., 2018)
Algorithm Hyperparameter
SVM Kernel type = Radial Basis Function
Gamma = 0.7 Penalty value = 10
CART Cross validation factor = 5
Max depth = 10
Minimum leaf population = 5 Minimum split population = 10
RF Number of trees = 30
Table 4. Overall, producer’s and user’s accuracies for oil palm class of each state and Peninsular Malaysia.
State Johor Kedah Kelantan Melaka Negeri
Sembilan
Pahang Pulau Pinang
Perak Perlis Selangor Terengganu Peninsular Malaysia
OA (%) 89.23 87.85 86.06 85.16 77.57 80.84 89.74 91.30 86.75 89.88 87.10 86.50
PA (%) 84.62 100.00 89.66 92.00 68.75 80.49 93.10 81.82 85.71 92.45 87.50 86.92
RF UA (%) 89.19 84.44 86.67 74.19 75.86 70.21 96.43 90.00 75.00 87.50 80.77 82.75
OA (%) 82.74 86.74 73.94 87.50 78.04 76.64 80.13 82.61 69.88 78.75 83.87 80.08
PA (%) 84.62 92.11 75.86 96.00 71.88 85.37 65.52 81.82 85.71 73.58 91.67 82.19
CART
UA (%) 76.74 85.37 73.33 80.00 58.97 70.00 65.52 81.82 50.00 88.64 59.46 71.82
OA (%) 89.38 97.35 98.18 88.28 88.32 81.28 96.15 97.10 97.59 97.62 93.55 93.16
PA (%) 89.74 97.22 100.00 92.00 87.50 89.19 96.55 90.91 100.00 98.11 87.50 93.52
SVM UA (%) 85.37 97.22 93.55 82.14 80.00 76.74 96.55 90.91 100.00 96.30 91.30 90.01
Note: OA: Overall accuracy (7 classes); PA: Producer’s accuracy (oil palm); UA: User’s accuracy (oil palm)
Oil palm area (ha)
RF CART SVM
State
MPOB Classified Difference Classified Difference Classified Difference
Johor 748860 799142 50282 752133 3273 782282 33422
Kedah 87538 147744 60206 157801 70263 170060 82522
Kelantan 158310 126177 -32133 183388 25078 106969 -51341
Melaka 57372 45768 -11604 42186 -15186 46021 -11351
Negeri Sembilan 184815 184325 -490 195733 10918 194311 9496
Pahang 741495 720745 -20750 802325 60830 717739 -23756
Pulau Pinang 13563 13146 -417 16039 2476 16572 3009
Perak 406469 392518 -13951 445041 38572 528448 121979
Perlis 660 1779 1119 3760 3100 4789 4129
Selangor 137783 196807 59024 195375 57592 194506 56723
Terengganu 171548 167136 -4412 211977 40429 162737 -8811
Table 6. Contingency table.